Lightweight insulator target detection method and system based on improved SSD algorithm

Abstract

The invention discloses a lightweight insulator target detection method and system based on an improved SSD algorithm, and the method comprises the steps: constructing an SSD network model, and carrying out the lightweight processing of the SSD network model, and obtaining a target SSD network model; performing model optimization on the target SSD network model according to a preset optimization rule to obtain a lightweight insulator target detection model based on an improved SSD algorithm; and inputting a real-time image containing an insulator into the lightweight insulator target detection model based on the improved SSD algorithm, wherein the lightweight insulator target detection model based on the improved SSD algorithm outputs a detection result corresponding to the real-time image. By analyzing the responsiveness of each characteristic path in the convolution module, weak response channels which contribute little to the detection performance are removed, and the parameter quantity and the calculation quantity of the model are remarkably reduced. Not only is the requirement for computing resources reduced, but also the hardware cost is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of insulator detection, and in particular relates to a lightweight insulator target detection method and system based on an improved SSD algorithm. Background Art

[0002] Insulators in power systems are exposed to high voltages for extended periods of time, and they can fail due to manufacturing defects, improper installation, environmental factors, or aging. Once a failure occurs, it not only impacts the normal operation of the power system but can also pose safety hazards such as fires and explosions. Therefore, regular, targeted inspections of insulators are crucial to identifying and addressing potential issues, ensuring safe power system operation.

[0003] The continuous expansion of power systems places higher demands on equipment maintenance and management. Traditional inspection methods are inefficient and struggle to achieve comprehensive coverage. Object detection technology enables remote monitoring of insulators, improving inspection efficiency and reducing maintenance costs. It also enables rapid response to abnormal situations, reducing power outages caused by equipment failures.

[0004] In recent years, with the development of computer vision and artificial intelligence technologies, object detection technology has become increasingly widely used in power systems. Using high-definition cameras to capture images of insulators, combined with deep learning algorithms, it is possible to automatically identify insulator conditions, including but not limited to surface damage, corrosion, and foreign matter adhesion, significantly improving detection accuracy and reliability. However, these often rely on complex deep learning models, such as Faster R-CNN and YOLOv3. While these models offer high detection accuracy, they are computationally intensive and require significant computing resources. Summary of the Invention

[0005] The present invention provides a lightweight insulator target detection method and system based on an improved SSD algorithm, which is used to solve the technical problem that the existing model has high detection accuracy but large calculation amount and requires high computing resources.

[0006] In a first aspect, the present invention provides a lightweight insulator target detection method based on an improved SSD algorithm, comprising: Construct an SSD network model and perform lightweight processing on the SSD network model to obtain a target SSD network model, wherein the lightweight processing is specifically as follows: When the mAP evaluation index of the SSD network model is at When the range is , the optimal weight of the SSD network model is saved; according to the preset optimization rate Perform feature path optimization on the SSD network model configured with optimal weights; The target SSD network model is optimized according to preset optimization rules to obtain a lightweight insulator target detection model based on the improved SSD algorithm, wherein the model optimization is specifically as follows: Obtain module optimization values ​​based on characteristic pathway responsiveness; Optimize the value of the module Perform module optimization on the first O associated convolution modules, and optimize the 3×3×C convolution kernel to a 1×1×C convolution kernel. The optimization formula of the convolution kernel is: , Where, is the optimized 1×1×C convolution kernel, is a 3×3×C convolution kernel, is the height of the convolution kernel, is the width of the convolution kernel, is the number of channels of the convolution kernel; A real-time image containing an insulator is input into the lightweight insulator target detection model based on the improved SSD algorithm, and the lightweight insulator target detection model based on the improved SSD algorithm outputs a detection result corresponding to the real-time image.

[0007] In a second aspect, the present invention provides a lightweight insulator target detection system based on an improved SSD algorithm, comprising: The construction module is configured to construct an SSD network model and perform lightweight processing on the SSD network model to obtain a target SSD network model, wherein the lightweight processing is specifically: When the mAP evaluation index of the SSD network model is at When the range is , the optimal weight of the SSD network model is saved; according to the preset optimization rate Perform feature path optimization on the SSD network model configured with optimal weights; The optimization module is configured to optimize the target SSD network model according to a preset optimization rule to obtain a lightweight insulator target detection model based on the improved SSD algorithm, wherein the model optimization is specifically as follows: Obtain module optimization values ​​based on characteristic pathway responsiveness; Optimize the value of the module Perform module optimization on the first O associated convolution modules, and optimize the 3×3×C convolution kernel to a 1×1×C convolution kernel. The optimization formula of the convolution kernel is: , Where, is the optimized 1×1×C convolution kernel, is a 3×3×C convolution kernel, is the height of the convolution kernel, is the width of the convolution kernel, is the number of channels of the convolution kernel; An output module is configured to input a real-time image containing an insulator into the lightweight insulator target detection model based on the improved SSD algorithm, and the lightweight insulator target detection model based on the improved SSD algorithm outputs a detection result corresponding to the real-time image.

[0008] According to a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the lightweight insulator target detection method based on the improved SSD algorithm of any embodiment of the present invention.

[0009] In a fourth aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon. When the program instructions are executed by a processor, the processor executes the steps of the lightweight insulator target detection method based on the improved SSD algorithm of any embodiment of the present invention.

[0010] The lightweight insulator target detection method and system based on the improved SSD algorithm proposed in this application analyzes the responsiveness of each feature path in the convolutional module and removes weakly responsive channels that contribute little to detection performance, significantly reducing the model's parameter count and computational complexity. This not only reduces the demand for computing resources but also reduces hardware costs. Module optimization technology: Optimizing each module of the model reduces redundant computations, further reducing computational complexity. These optimization measures enable the model to run efficiently on resource-constrained devices, such as embedded systems and mobile devices. Lightweight model: Through feature path optimization and module optimization, the model's computational complexity is significantly reduced, enabling real-time detection on embedded systems and mobile devices. This is particularly important for the rapid response requirements of power systems, enabling timely fault detection and resolution, and reducing power outages caused by equipment failures. Efficient inference: The lightweight model offers faster inference speed and can quickly process high-resolution images, ensuring timely and accurate detection results. High resource utilization: The lightweight model requires fewer computing resources, reducing hardware and maintenance costs. Easy deployment: The lightweight model can run efficiently on a variety of devices, improving system flexibility and scalability while reducing deployment and maintenance complexity. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0012] Figure 1 A flowchart of a lightweight insulator target detection method based on an improved SSD algorithm provided by one embodiment of the present invention; Figure 2 A structural block diagram of a lightweight insulator target detection system based on an improved SSD algorithm provided by one embodiment of the present invention; Figure 3 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0013] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0014] See also Figure 1 , which shows a flowchart of a lightweight insulator target detection method based on an improved SSD algorithm of the present application.

[0015] like Figure 1 As shown, a lightweight insulator target detection method based on the improved SSD algorithm specifically includes the following steps: Step S101: construct an SSD network model, and perform lightweight processing on the SSD network model to obtain a target SSD network model.

[0016] In this step, the convolutional neural network of VGG16 is used as the base network to extract high-level features of the input image. The output of the base network is a feature map that contains high-level semantic information of the image.

[0017] To detect objects of different scales, the SSD network model adds multiple convolutional layers to the base network to generate feature maps of different scales. Each feature map corresponds to an object detection of a different scale.

[0018] The SSD network model predefines a number of anchor boxes for each location in each feature map. These default boxes have different scales and aspect ratios to capture objects of different sizes and shapes. The number and shape of the default boxes are predefined.

[0019] For each default box on each feature map, the SSD network model uses two convolutional layers to make predictions: Classification Layer: Predicts the probability that each default box belongs to each category. The output dimension is [num_boxes, num_classes], where num_boxes is the number of default boxes on the feature map, and num_classes is the number of categories (including background classes).

[0020] Regression Layer: Predicts the offset of each default box, which is used to adjust the position and size of the default box. The output dimension is [num_boxes, 4], which represents the four coordinate offsets (dx, dy, dw, dh) of each default box.

[0021] Non-Maximum Suppression (NMS) During the prediction phase, the SSD network model performs non-maximum suppression (NMS) on the prediction results of each default box. NMS is used to remove overlapping detection boxes and retain the most likely detection results. The specific steps are as follows: For each category, select the detection box with the highest confidence.

[0022] Remove other detection boxes whose overlap (IoU) with the detection box exceeds a certain threshold.

[0023] Repeat the above steps until all detection boxes have been processed.

[0024] The SSD network model uses a multi-task loss function, which is expressed as: , Where, is the number of matching anchor boxes, is the balance factor, For the The classification loss function of anchor boxes, For the The regression loss function of the anchor box.

[0025] It should be noted that the specific network structure of the SSD network model is as follows: Feature encoding and feature decoding structure: The first to ninth convolutional modules are sequentially connected to the corresponding first to eighth downsampling convolutional modules. Specifically, starting from the first convolutional module, it passes through the first downsampling convolutional module in sequence, then connects to the second convolutional module, continues through the second downsampling convolutional module, and so on, until the last eighth downsampling convolutional module is connected to the ninth convolutional module, forming a gradually deepening feature extraction process.

[0026] Detection Structure: In the detection section, the ninth convolutional module is directly connected to the first convolutional module, the eighth convolutional module is connected to the second detection head, the seventh convolutional module is connected to the third detection head, the sixth convolutional module is connected to the fourth detection head, and the fifth convolutional module is connected to the fifth detection head. Each detection head (first through fifth) is connected to a non-maximum suppression module, which integrates information from each detection head and outputs accurate detection results by suppressing non-maximum responses.

[0027] Convolutional module: The input feature map is first concatenated with a convolution operation. The output of the convolution operation (with a kernel size of 3×3, a stride of 1, and padding) is then concatenated with batch normalization. The output of batch normalization is then concatenated with a ReLU activation function. Finally, the input feature map is added to the output of the ReLU activation function after the above processing to produce the final output feature map.

[0028] Downsampling Convolutional Module: In the downsampling convolutional module, the input feature map is first convolved with a 3×3 convolution kernel, a stride of 2, and padding. The output of the convolution is then convolved with batch normalization, which is then convolved with a ReLU activation function. The output of the ReLU activation function ultimately becomes the output feature map of the module.

[0029] Specifically, the lightweight processing of the SSD network model is as follows: When the mAP evaluation index of the SSD network model is at When the range is , the optimal weight of the SSD network model is saved; according to the preset optimization rate Perform feature path optimization on the SSD network model with optimal weights. Specifically: S1. Based on the method of obtaining the responsiveness of a single feature path in a single convolutional module, the responsiveness of all feature paths of the convolutional module feature map in the SSD network model is obtained; S2. Sort all characteristic pathway responsiveness in descending order, and place the sort position at The feature map channel and convolution kernel associated with the subsequent feature path responsiveness are respectively called weak response feature map channel and weak response convolution kernel, and the convolution kernel channel associated with the weak response feature map channel is called weak response convolution kernel channel; S3, based on the preset removal strategy, remove the weak response feature map channel, weak response convolution kernel and weak response convolution kernel channel, wherein the removal strategy is: remove The weak response feature map channel of The weak response convolution kernel channel of the convolution is The previous convolution module Weakly responsive convolution kernel or downsampling convolution module The weak response convolution kernel is removed, where The input image is Time Convolutional modules The input feature map of .

[0030] S4. After optimizing the feature map of a single convolution module, the SSD network model with optimized feature paths needs to be fine-tuned. When the performance index of the model tends to be stable, that is, the fluctuation of the mAP evaluation index in three consecutive training cycles remains at Only when it is within the range can the next feature path optimization operation be performed. The default value is 0.005; S5. Perform the operations of S2 to S4 on the first convolution module to the fifth convolution module and the first downsampling convolution module to the fourth downsampling convolution module in sequence to complete the feature path optimization.

[0031] Get the convolution module feature path responsiveness specified in the SSD network model: First, the image As the input of the SSD network model after configuring the optimal weight, we get 、 And the final output feature map of the SSD network model . Use replace As the first Convolutional modules The input feature map is obtained Convolutional modules The output feature map of And the final output feature map of the SSD network model .

[0032] The expression for calculating the characteristic pathway responsiveness is: , Where, For the Convolutional modules No. Characteristic pathway responsiveness, is the total number of test set images in the dataset, For the convolution module The number of vertical pixels of the input and output feature maps, For the convolution module The number of horizontal pixels of the input and output feature maps, is the output feature map response weight, is the intermediate feature map response weight, The input image for the SSD network model is Time Convolutional modules The input feature map is composed of Replace with The first Convolutional modules The output feature map of The input image is Time Convolutional modules The output feature map of After configuring the optimal weight, the SSD network model will As input As the first Convolutional modules The final output feature map obtained when for Input the SSD network model with the optimal weight configuration to obtain the output feature map. The input image is Time Convolutional modules The input feature map of is the test factor.

[0033] Step S102, optimizing the target SSD network model according to preset optimization rules to obtain a lightweight insulator target detection model based on the improved SSD algorithm. In this step, the model optimization is specifically as follows: The module optimization value is obtained according to the characteristic path responsiveness. The module optimization value calculation formula of the convolution module is: , Where, Optimize values ​​for modules, is the total number of test set images in the dataset, For the Convolutional modules No. Characteristic pathway responsiveness, For the convolution module The number of vertical pixels of the input and output feature maps, For the convolution module The number of horizontal pixels of the input and output feature maps, The input image for the SSD network model is Time Convolutional modules The input feature map is composed of Replace with The first Convolutional modules The output feature map of The input image is Time Convolutional modules The output feature map of After configuring the optimal weight, the SSD network model will As input As the first Convolutional modules The final output feature map obtained when for Input the SSD network model with the optimal weight configuration to obtain the output feature map. The input image is Time Convolutional modules The input feature map of Is the test factor. Optimize the value of the module Perform module optimization on the first O associated convolution modules, and optimize the 3×3×C convolution kernel to a 1×1×C convolution kernel. The optimization formula of the convolution kernel is: , Where, is the optimized 1×1×C convolution kernel, is a 3×3×C convolution kernel, is the height of the convolution kernel, is the width of the convolution kernel, is the number of channels of the convolution kernel; Step S103 : inputting a real-time image containing the insulator into the lightweight insulator target detection model based on the improved SSD algorithm, and the lightweight insulator target detection model based on the improved SSD algorithm outputs a detection result corresponding to the real-time image.

[0034] In summary, the method of this application significantly reduces the number of model parameters and computational complexity by analyzing the responsiveness of each feature path in the convolutional module and removing weakly responsive channels that contribute little to detection performance. This not only reduces the demand for computing resources but also reduces hardware costs. Module optimization technology: Optimizing each module of the model reduces redundant computations, further reducing computational complexity. These optimization measures enable the model to run efficiently on resource-constrained devices, such as embedded systems and mobile devices. Lightweight model: Through feature path optimization and module optimization, the model's computational complexity is significantly reduced, enabling real-time detection on embedded systems and mobile devices. This is particularly important for the rapid response requirements of power systems, enabling timely fault detection and resolution, and reducing power outages caused by equipment failures. Efficient inference: The lightweight model offers faster inference speed and can quickly process high-resolution images, ensuring timely and accurate detection results. High resource utilization: The lightweight model requires fewer computing resources, reducing hardware and maintenance costs. Ease of deployment: The lightweight model can run efficiently on a variety of devices, improving system flexibility and scalability while reducing deployment and maintenance complexity.

[0035] See also Figure 2 , which shows a structural block diagram of a lightweight insulator target detection system based on an improved SSD algorithm of the present application.

[0036] like Figure 2 As shown, the lightweight insulator target detection system 200 based on the improved SSD algorithm includes a construction module 210 , an optimization module 220 and an output module 230 .

[0037] The construction module 210 is configured to construct an SSD network model and perform lightweight processing on the SSD network model to obtain a target SSD network model, wherein the lightweight processing is specifically as follows: When the mAP evaluation index of the SSD network model is at When the range is , the optimal weight of the SSD network model is saved; according to the preset optimization rate Perform feature path optimization on the SSD network model configured with optimal weights; The optimization module 220 is configured to optimize the target SSD network model according to a preset optimization rule to obtain a lightweight insulator target detection model based on the improved SSD algorithm, wherein the model optimization is specifically as follows: Obtain module optimization values ​​based on characteristic pathway responsiveness; Optimize the value of the module Perform module optimization on the first O associated convolution modules, and optimize the 3×3×C convolution kernel to a 1×1×C convolution kernel. The optimization formula of the convolution kernel is: , Where, is the optimized 1×1×C convolution kernel, is a 3×3×C convolution kernel, is the height of the convolution kernel, is the width of the convolution kernel, is the number of channels of the convolution kernel; The output module 230 is configured to input a real-time image containing an insulator into the lightweight insulator target detection model based on the improved SSD algorithm, and the lightweight insulator target detection model based on the improved SSD algorithm outputs a detection result corresponding to the real-time image.

[0038] It should be understood that Figure 2 Modules and references documented in Figure 1 Therefore, the operations and features described above for the method and the corresponding technical effects also apply to Figure 2 The modules in it will not be described in detail here.

[0039] In other embodiments, embodiments of the present invention further provide a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor is caused to execute the lightweight insulator target detection method based on the improved SSD algorithm in any of the above method embodiments; As an embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, and the computer-executable instructions are configured as follows: Construct an SSD network model and perform lightweight processing on the SSD network model to obtain a target SSD network model, wherein the lightweight processing is specifically as follows: When the mAP evaluation index of the SSD network model is at When the range is , the optimal weight of the SSD network model is saved; according to the preset optimization rate Perform feature path optimization on the SSD network model configured with optimal weights; The target SSD network model is optimized according to preset optimization rules to obtain a lightweight insulator target detection model based on the improved SSD algorithm, wherein the model optimization is specifically as follows: Obtain module optimization values ​​based on characteristic pathway responsiveness; Optimize the value of the module Perform module optimization on the first O associated convolution modules, and optimize the 3×3×C convolution kernel to a 1×1×C convolution kernel. The optimization formula of the convolution kernel is: , Where, is the optimized 1×1×C convolution kernel, is a 3×3×C convolution kernel, is the height of the convolution kernel, is the width of the convolution kernel, is the number of channels of the convolution kernel; A real-time image containing an insulator is input into the lightweight insulator target detection model based on the improved SSD algorithm, and the lightweight insulator target detection model based on the improved SSD algorithm outputs a detection result corresponding to the real-time image.

[0040] The computer-readable storage medium may include a program storage area and a data storage area. The program storage area may store an operating system and application programs required for at least one function; the data storage area may store data generated based on the use of the lightweight insulator target detection system based on the improved SSD algorithm. Furthermore, the computer-readable storage medium may include high-speed random access memory and may also include storage, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include storage remotely located from the processor. Such remote storage may be connected to the lightweight insulator target detection system based on the improved SSD algorithm via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0041] Figure 3 Schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. Figure 3 As shown, the device includes: a processor 310 and a memory 320. The electronic device may also include: an input device 330 and an output device 340. The processor 310, the memory 320, the input device 330 and the output device 340 may be connected via a bus or other means. Figure 3 The example uses a bus connection. Memory 320 is the aforementioned computer-readable storage medium. Processor 310 executes the various server functional applications and data processing by running the non-volatile software programs, instructions, and modules stored in memory 320, thereby implementing the lightweight insulator target detection method based on the improved SSD algorithm described in the aforementioned method embodiment. Input device 330 can receive input digital or character information and generate key signal input related to user settings and function control of the lightweight insulator target detection system based on the improved SSD algorithm. Output device 340 can include a display device such as a display screen.

[0042] The electronic device can execute the method provided by the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided by the embodiment of the present invention.

[0043] As an embodiment, the electronic device is applied to a lightweight insulator target detection system based on an improved SSD algorithm and is used for a client, and includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: Construct an SSD network model and perform lightweight processing on the SSD network model to obtain a target SSD network model, wherein the lightweight processing is specifically as follows: When the mAP evaluation index of the SSD network model is at When the range is , the optimal weight of the SSD network model is saved; according to the preset optimization rate Perform feature path optimization on the SSD network model configured with optimal weights; The target SSD network model is optimized according to preset optimization rules to obtain a lightweight insulator target detection model based on the improved SSD algorithm, wherein the model optimization is specifically as follows: Obtain module optimization values ​​based on characteristic pathway responsiveness; Optimize the value of the module Perform module optimization on the first O associated convolution modules, and optimize the 3×3×C convolution kernel to a 1×1×C convolution kernel. The optimization formula of the convolution kernel is: , Where, is the optimized 1×1×C convolution kernel, is a 3×3×C convolution kernel, is the height of the convolution kernel, is the width of the convolution kernel, is the number of channels of the convolution kernel; A real-time image containing an insulator is input into the lightweight insulator target detection model based on the improved SSD algorithm, and the lightweight insulator target detection model based on the improved SSD algorithm outputs a detection result corresponding to the real-time image.

[0044] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units. That is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0045] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A lightweight insulator target detection method based on an improved SSD algorithm, characterized in that: include: An SSD network model is constructed, and the SSD network model is lightweight processed to obtain a target SSD network model, wherein the lightweight processing is specifically as follows: When the mAP evaluation index of the SSD network model is at When the optimal weight of the SSD network model is saved, the optimal weight of the SSD network model is saved; according to the preset optimization rate Perform feature path optimization on the SSD network model configured with optimal weights; The target SSD network model is optimized according to the preset optimization rules to obtain a lightweight insulator target detection model based on the improved SSD algorithm, wherein the model optimization is specifically as follows: Obtain module optimization values ​​based on characteristic pathway responsiveness; Optimize the value of the module The first O associated convolution modules are optimized to optimize the 3×3×C convolution kernel to a 1×1×C convolution kernel, where the optimization formula of the convolution kernel is: ， In the formula, is the optimized 1×1×C convolution kernel, is a 3×3×C convolution kernel, is the height of the convolution kernel, is the width of the convolution kernel, is the number of channels of the convolution kernel; A real-time image containing an insulator is input into the lightweight insulator target detection model based on the improved SSD algorithm, and the lightweight insulator target detection model based on the improved SSD algorithm outputs a detection result corresponding to the real-time image.

2. The lightweight insulator target detection method based on the improved SSD algorithm according to claim 1 is characterized in that: The expression of the multi-task loss function of the SSD network model is: ， In the formula, is the number of matching anchor boxes, is the balance factor, For the The classification loss function of anchor boxes is: For the The regression loss function of the anchor box.

3. The lightweight insulator target detection method based on the improved SSD algorithm according to claim 1 is characterized in that: The module optimization value calculation formula of the convolution module is: ， In the formula, Optimize values ​​for modules, is the total number of test set images in the dataset, For the Convolutional modules No. Characteristic pathway responsiveness, For the convolution module The number of vertical pixels of the input and output feature maps, For the convolution module The number of horizontal pixels of the input and output feature maps, The input image for the SSD network model is Time Convolutional modules The input feature map is composed of Replace with The first Convolutional modules The output feature map of The input image is Time Convolutional modules The output feature map of After configuring the optimal weights, the SSD network model will As input As the Convolutional modules The final output feature map obtained when for Input the SSD network model with the optimal weights to obtain the output feature map. The input image is Time Convolutional modules The input feature map is is the test factor.

4. The lightweight insulator target detection method based on the improved SSD algorithm according to claim 1 is characterized in that: The preset optimization rate Performing feature path optimization on the SSD network model with optimal weights includes: Based on the method of obtaining the responsiveness of a single feature path in a single convolutional module, the responsiveness of all feature paths of the convolutional module feature map in the SSD network model is obtained; Sort all characteristic pathway responsiveness in descending order, and place the sort position at The feature map channel and convolution kernel associated with the feature path responsiveness are respectively called weak response feature map channel and weak response convolution kernel, and the convolution kernel channel associated with the weak response feature map channel is called weak response convolution kernel channel; Based on the preset removal strategy, the weak response feature map channel, the weak response convolution kernel and the weak response convolution kernel channel are removed, wherein the removal strategy is: remove The weak response feature map channel of The weak response convolution kernel channel of the convolution The previous convolution module Weakly responsive convolution kernel or downsampling convolution module The weak response convolution kernel of is removed, where The input image is Time Convolutional modules The input feature map of .

5. The lightweight insulator target detection method based on the improved SSD algorithm according to claim 4 is characterized in that: in, The expression for calculating the characteristic pathway responsiveness is: ， In the formula, For the Convolutional modules No. Characteristic pathway responsiveness, is the total number of test set images in the dataset, For the convolution module The number of vertical pixels of the input and output feature maps, For the convolution module The number of horizontal pixels of the input and output feature maps, is the output feature map response weight, is the intermediate feature map response weight, The input image for the SSD network model is Time Convolutional modules The input feature map is composed of Replace with The first Convolutional modules The output feature map of The input image is Time Convolutional modules The output feature map of After configuring the optimal weights, the SSD network model will As input As the Convolutional modules The final output feature map obtained when for Input the SSD network model with the optimal weights to obtain the output feature map. The input image is Time Convolutional modules The input feature map is is the test factor.

6. A lightweight insulator target detection system based on an improved SSD algorithm, characterized in that: include: The construction module is configured to construct an SSD network model and perform lightweight processing on the SSD network model to obtain a target SSD network model, wherein the lightweight processing is specifically: When the mAP evaluation index of the SSD network model is at When the optimal weight of the SSD network model is saved, the optimal weight of the SSD network model is saved; according to the preset optimization rate Perform feature path optimization on the SSD network model configured with optimal weights; The optimization module is configured to optimize the target SSD network model according to a preset optimization rule to obtain a lightweight insulator target detection model based on the improved SSD algorithm, wherein the model optimization is specifically as follows: Obtain module optimization values ​​based on characteristic pathway responsiveness; Optimize the value of the module The first O associated convolution modules are optimized to optimize the 3×3×C convolution kernel to a 1×1×C convolution kernel, where the optimization formula of the convolution kernel is: ， In the formula, is the optimized 1×1×C convolution kernel, is a 3×3×C convolution kernel, is the height of the convolution kernel, is the width of the convolution kernel, is the number of channels of the convolution kernel; The output module is configured to input a real-time image containing an insulator into the lightweight insulator target detection model based on the improved SSD algorithm, and the lightweight insulator target detection model based on the improved SSD algorithm outputs a detection result corresponding to the real-time image.

7. An electronic device, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images